svn commit: r211130 - head/libexec/rtld-elf/mips

[Bruce Evans]

On Tue, 10 Aug 2010, Stefan Farfeleder wrote:

> On Tue, Aug 10, 2010 at 12:55:38PM +0530, Jayachandran C. wrote:
>>
>> I think there is a problem in  sys/mips/include/_endian.h
>> --
>> #define __bswap16(x)    (__uint16_t)(__is_constant(x) ?         \
>>         __bswap16_const((__uint16_t)x) :  __bswap16_var((__uint16_t)x))
>> #define __bswap32(x)    (__uint32_t)(__is_constant(x) ?         \
>>         __bswap32_const((__uint32_t)x) :  __bswap32_var((__uint32_t)x))
>> #define __bswap64(x)    (__uint64_t)(__is_constant(x) ?         \
>>         __bswap64_const((__uint64_t)x) :  __bswap64_var((__uint64_t)x))
>> --
>>
>> I'm not sure why the cast is needed, but we should have a braces
>> around x, unless I'm completely mistaken.

And not around x in callers.

> I agree. And around the entire expression too.

I agree.  But I've never seen a case where foo(x) needs to be (foo(x))
(where foo is not a macro).  '()' has highest precedence for expressions
(not sure about for parameter lists) together with `[]' `->' and '.', and
it is hard to think of an expression with any of the latter close enough
to foo(x) to cause problems.

Bruce